Forensic detection of processing operator chains: recovering the history of filtered JPEG images
Powerful image editing software is nowadays capable of creating sophisticated and visually compelling fake photographs, thus posing serious issues to the trustworthiness of digital contents as a true representation of reality. Digital image forensics has emerged to help regain some trust in digital imagesby providing valuable aids in learning the history of an image. Unfortunately, in real scenarios its application is limited, since multiple processing operators are likely to be applied, which alters the characteristic footprints exploited by current forensic tools.
In this work, we develop a novel forensic technique that is able to detect chains of operators applied to an image. In particular, we study the combination of JPEG compression and full-frame linear filtering, and derive an accurate mathematical framework to fully characterize the probabilistic distributions of the Discrete Cosine Transform (DCT) coefficients of the quantized and filtered image. We then exploit such knowledge to define a set of features from the DCT distribution and build an effective classifier able to jointly disclose the quality factor of the applied compression and the filter kernel. Extensive experimental analysis illustrates the efficiency and versatility of the proposed approach, which effectively overcomes the state-of-the-art.